Newton reads a 2D mammogram (left) through a staged pipeline — modality → implant → density → structure — localizes the suspicious cluster, then rebuilds it on a real anatomical breast (Visible Human, public domain) as a coloured mass with an approximate surgical margin. Attention cluster is illustrative; depth (CC+MLO) comes with the trained detector. Segmentation is now validated separately: given a lesion’s location as a region prompt (MedSAM protocol), Newton outlines it at 0.847 mean Dice across 370 held-out discrete lesions from patients never seen in training (median 0.897; masses 0.879, calcifications 0.826; no benign/malignant bias). That measures outlining precision, not detection — finding the lesion unaided on a full mammogram is the next stage and is not built yet. Research demonstration, not a clinical diagnosis.